!4274 add quantize resize bilinear and nearest neighbor

Merge pull request !4274 from zhaozhenlong/lite/op/int8/resize

mindspore/lite/src/ops/pad.cc

@@ -44,8 +44,10 @@ int Pad::InferShape(std::vector<tensor::Tensor *> inputs, std::vector<tensor::Te
   }
   auto input_shape = input->shape();
   std::vector<int> output_shape;
+  MS_ASSERT(input->shape().size() <= kInputRank);
   for (size_t i = 0; i < input_shape.size(); i++) {
-    auto shape = input_shape[i] + (*paddings)[2 * i] + (*paddings)[2 * i + 1];
+    auto paddings_index = i + kInputRank - input_shape.size();
+    auto shape = input_shape[i] + (*paddings)[2 * paddings_index] + (*paddings)[2 * paddings_index + 1];
     output_shape.push_back(shape);
   }
 

mindspore/lite/src/ops/resize.cc

@@ -30,6 +30,8 @@ int Resize::InferShape(std::vector<tensor::Tensor *> inputs_, std::vector<tensor
   if (input == nullptr) {
     return RET_NULL_PTR;
   }
+  MS_ASSERT(input->shape().size() == kInputRank);
+
   auto output = outputs_.front();
   if (output == nullptr) {
     return RET_NULL_PTR;

mindspore/lite/src/populate_parameter.cc

@@ -57,7 +57,7 @@
 #include "src/runtime/kernel/arm/nnacl/unique.h"
 #include "src/runtime/kernel/arm/nnacl/scale.h"
 #include "src/runtime/kernel/arm/nnacl/fp32/gatherNd.h"
-#include "src/runtime/kernel/arm/nnacl/resize.h"
+#include "src/runtime/kernel/arm/nnacl/resize_parameter.h"
 #include "src/runtime/kernel/arm/nnacl/scatter_nd.h"
 #include "src/runtime/kernel/arm/nnacl/batch_to_space.h"
 #include "src/runtime/kernel/arm/nnacl/fp32/crop.h"
@@ -1023,7 +1023,7 @@ OpParameter *PopulateResizeParameter(const lite::Primitive *primitive) {
   }
   resize_param->op_parameter_.type_ = primitive->Type();
   auto param = primitive->Value()->value_as_Resize();
-  resize_param->method_ = param->method();
+  resize_param->method_ = static_cast<int>(param->method());
   resize_param->new_height_ = param->newHeight();
   resize_param->new_width_ = param->newWidth();
   resize_param->align_corners_ = param->alignCorners();

mindspore/lite/src/runtime/kernel/arm/base/resize_base.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <vector>
+#include "schema/model_generated.h"
+#include "src/kernel_registry.h"
+#include "include/errorcode.h"
+#include "src/runtime/kernel/arm/base/resize_base.h"
+#include "src/runtime/kernel/arm/fp32/resize.h"
+#include "src/runtime/kernel/arm/int8/resize_int8.h"
+using mindspore::lite::KernelRegistrar;
+using mindspore::lite::RET_ERROR;
+using mindspore::lite::RET_INVALID_OP_ATTR;
+using mindspore::lite::RET_NULL_PTR;
+using mindspore::lite::RET_OK;
+
+namespace mindspore::kernel {
+namespace {
+constexpr int kInputNum = 1;
+constexpr int kOutputNum = 1;
+constexpr int kRank = 4;
+}  // namespace
+
+int ResizeBaseCPUKernel::CheckParameters() {
+  auto parameter = reinterpret_cast<ResizeParameter *>(opParameter);
+  if (parameter == nullptr) {
+    MS_LOG(ERROR) << "cast ResizeParameter failed.";
+    return RET_NULL_PTR;
+  }
+  method_ = parameter->method_;
+  if (method_ != schema::ResizeMethod_BILINEAR && method_ != schema::ResizeMethod_NEAREST_NEIGHBOR) {
+    MS_LOG(ERROR) << "Resize method should be bilinear or nearest_neighbor, but got " << method_;
+    return RET_INVALID_OP_ATTR;
+  }
+  new_height_ = parameter->new_height_;
+  if (new_height_ < 1) {
+    MS_LOG(ERROR) << "Resize new_height should >= 1, but got " << new_height_;
+    return RET_INVALID_OP_ATTR;
+  }
+  new_width_ = parameter->new_width_;
+  if (new_width_ < 1) {
+    MS_LOG(ERROR) << "Resize new_width should >= 1, but got " << new_width_;
+    return RET_INVALID_OP_ATTR;
+  }
+  align_corners_ = parameter->align_corners_;
+  preserve_aspect_ratio = parameter->preserve_aspect_ratio_;
+  if (preserve_aspect_ratio) {
+    MS_LOG(ERROR) << "Resize currently not support preserve_aspect_ratio true";
+    return RET_ERROR;
+  }
+  return RET_OK;
+}
+
+int ResizeBaseCPUKernel::CheckInputsOuputs() {
+  if (inputs_.size() != kInputNum) {
+    MS_LOG(ERROR) << "Resize input num should be " << kInputNum << ", but got " << inputs_.size();
+    return RET_ERROR;
+  }
+  auto input = inputs_.at(0);
+  if (input == nullptr) {
+    return RET_NULL_PTR;
+  }
+  if (outputs_.size() != kOutputNum) {
+    MS_LOG(ERROR) << "Resize output num should be " << kOutputNum << ", but got " << outputs_.size();
+    return RET_ERROR;
+  }
+  auto output = outputs_.at(0);
+  if (output == nullptr) {
+    return RET_NULL_PTR;
+  }
+  return RET_OK;
+}
+
+int ResizeBaseCPUKernel::Init() {
+  auto ret = CheckParameters();
+  if (ret != RET_OK) {
+    return ret;
+  }
+  ret = CheckInputsOuputs();
+  if (ret != RET_OK) {
+    return ret;
+  }
+
+  auto input = inputs_.at(0);
+  auto input_shape = input->shape();
+  if (input_shape.size() != kRank) {
+    MS_LOG(ERROR) << "Resize op support input rank 4, got " << input_shape.size();
+    return RET_ERROR;
+  }
+
+  return RET_OK;
+}
+
+kernel::LiteKernel *CpuResizeFp32KernelCreator(const std::vector<lite::tensor::Tensor *> &inputs,
+                                               const std::vector<lite::tensor::Tensor *> &outputs,
+                                               OpParameter *opParameter, const lite::Context *ctx,
+                                               const kernel::KernelKey &desc, const lite::Primitive *primitive) {
+  if (opParameter == nullptr) {
+    MS_LOG(ERROR) << "Input opParameter is nullptr!";
+    return nullptr;
+  }
+  MS_ASSERT(desc.type == schema::PrimitiveType_Resize);
+  auto *kernel = new (std::nothrow) ResizeCPUKernel(opParameter, inputs, outputs, ctx, primitive);
+  if (kernel == nullptr) {
+    MS_LOG(ERROR) << "new ResizeCPUKernel fail!";
+    return nullptr;
+  }
+  auto ret = kernel->Init();
+  if (ret != RET_OK) {
+    delete kernel;
+    MS_LOG(ERROR) << "Init kernel failed, name: " << opParameter->name_ << ", type: "
+                  << schema::EnumNamePrimitiveType(static_cast<schema::PrimitiveType>(opParameter->type_));
+    return nullptr;
+  }
+
+  return kernel;
+}
+
+kernel::LiteKernel *CpuResizeInt8KernelCreator(const std::vector<lite::tensor::Tensor *> &inputs,
+                                               const std::vector<lite::tensor::Tensor *> &outputs,
+                                               OpParameter *opParameter, const lite::Context *ctx,
+                                               const kernel::KernelKey &desc, const lite::Primitive *primitive) {
+  if (opParameter == nullptr) {
+    MS_LOG(ERROR) << "Input opParameter is nullptr!";
+    return nullptr;
+  }
+  MS_ASSERT(desc.type == schema::PrimitiveType_Resize);
+  auto *kernel = new (std::nothrow) ResizeInt8CPUKernel(opParameter, inputs, outputs, ctx, primitive);
+  if (kernel == nullptr) {
+    MS_LOG(ERROR) << "new ResizeCPUKernel fail!";
+    return nullptr;
+  }
+  auto ret = kernel->Init();
+  if (ret != RET_OK) {
+    delete kernel;
+    MS_LOG(ERROR) << "Init kernel failed, name: " << opParameter->name_ << ", type: "
+                  << schema::EnumNamePrimitiveType(static_cast<schema::PrimitiveType>(opParameter->type_));
+    return nullptr;
+  }
+
+  return kernel;
+}
+
+REG_KERNEL(kCPU, kNumberTypeInt8, PrimitiveType_Resize, CpuResizeInt8KernelCreator)
+REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_Resize, CpuResizeFp32KernelCreator)
+}  // namespace mindspore::kernel

mindspore/lite/src/runtime/kernel/arm/base/resize_base.h

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_BASE_RESIZE_BASE_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_BASE_RESIZE_BASE_H_
+
+#include <vector>
+#include "src/lite_kernel.h"
+#include "nnacl/resize_parameter.h"
+
+using mindspore::schema::PrimitiveType_Resize;
+using mindspore::schema::ResizeMethod;
+
+namespace mindspore::kernel {
+class ResizeBaseCPUKernel : public LiteKernel {
+ public:
+  ResizeBaseCPUKernel(OpParameter *parameter, const std::vector<lite::tensor::Tensor *> &inputs,
+                  const std::vector<lite::tensor::Tensor *> &outputs, const lite::Context *ctx,
+                  const lite::Primitive *primitive)
+    : LiteKernel(parameter, inputs, outputs, ctx, primitive), context_(ctx) {}
+
+  ~ResizeBaseCPUKernel() = default;
+
+  int Init() override;
+  int ReSize() override { return 0; };
+
+ protected:
+  const lite::Context *context_;
+  int method_;
+  int64_t new_height_;
+  int64_t new_width_;
+  bool align_corners_;
+  bool preserve_aspect_ratio;
+
+ private:
+  int CheckParameters();
+  int CheckInputsOuputs();
+};
+}  // namespace mindspore::kernel
+
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_BASE_RESIZE_BASE_H_

mindspore/lite/src/runtime/kernel/arm/fp32/resize.cc

@@ -15,136 +15,33 @@
  */
 
 #include "src/runtime/kernel/arm/fp32/resize.h"
-#include <vector>
 #include "schema/model_generated.h"
-#include "src/kernel_registry.h"
-#include "src/runtime/kernel/arm/nnacl/resize.h"
-#include "src/runtime/kernel/arm/nnacl/pack.h"
+#include "nnacl/fp32/resize.h"
 #include "include/errorcode.h"
 #include "src/runtime/runtime_api.h"
 
 using mindspore::kernel::KERNEL_ARCH::kCPU;
-using mindspore::lite::KernelRegistrar;
 using mindspore::lite::RET_ERROR;
 using mindspore::lite::RET_INVALID_OP_ATTR;
 using mindspore::lite::RET_NULL_PTR;
 using mindspore::lite::RET_OK;
 
 namespace mindspore::kernel {
-namespace {
-constexpr int kInputNum = 1;
-constexpr int kOutputNum = 1;
-constexpr int kRank = 4;
-}  // namespace
-
-int ResizeCPUKernel::CheckParameters() {
-  auto parameter = reinterpret_cast<ResizeParameter *>(opParameter);
-  if (parameter == nullptr) {
-    MS_LOG(ERROR) << "cast ResizeParameter failed.";
-    return RET_NULL_PTR;
-  }
-  method_ = parameter->method_;
-  if (method_ != schema::ResizeMethod_BILINEAR && method_ != schema::ResizeMethod_NEAREST_NEIGHBOR) {
-    MS_LOG(ERROR) << "Resize method should be bilinear or nearest_neighbor, but got " << method_;
-    return RET_INVALID_OP_ATTR;
-  }
-  new_height_ = parameter->new_height_;
-  if (new_height_ < 1) {
-    MS_LOG(ERROR) << "Resize new_height should >= 1, but got " << new_height_;
-    return RET_INVALID_OP_ATTR;
-  }
-  new_width_ = parameter->new_width_;
-  if (new_width_ < 1) {
-    MS_LOG(ERROR) << "Resize new_width should >= 1, but got " << new_width_;
-    return RET_INVALID_OP_ATTR;
-  }
-  align_corners_ = parameter->align_corners_;
-  preserve_aspect_ratio = parameter->preserve_aspect_ratio_;
-  if (preserve_aspect_ratio) {
-    MS_LOG(ERROR) << "Resize currently not support preserve_aspect_ratio true";
-    return RET_ERROR;
-  }
-  return RET_OK;
-}
-
-int ResizeCPUKernel::CheckInputsOuputs() {
-  if (inputs_.size() != kInputNum) {
-    MS_LOG(ERROR) << "Resize input num should be " << kInputNum << ", but got " << inputs_.size();
-    return RET_ERROR;
-  }
-  auto input = inputs_.at(0);
-  if (input == nullptr) {
-    return RET_NULL_PTR;
-  }
-  if (outputs_.size() != kOutputNum) {
-    MS_LOG(ERROR) << "Resize output num should be " << kOutputNum << ", but got " << outputs_.size();
-    return RET_ERROR;
-  }
-  auto output = outputs_.at(0);
-  if (output == nullptr) {
-    return RET_NULL_PTR;
-  }
-  return RET_OK;
-}
-
 int ResizeCPUKernel::Init() {
-  if (context_->infer_shape_interrupt_ && !context_->running_) {
-    SetNeedReInit();
-    return RET_OK;
-  }
-  auto ret = CheckParameters();
-  if (ret != RET_OK) {
-    return ret;
-  }
-  ret = CheckInputsOuputs();
+  auto ret = ResizeBaseCPUKernel::Init();
   if (ret != RET_OK) {
     return ret;
   }
-
-  auto output = outputs_.at(0);
-  auto input = inputs_.at(0);
-  auto input_shape = input->shape();
-  if (input_shape.size() != kRank) {
-    return RET_ERROR;
-  }
-  schema::Format execute_format;
-  size_t exec_input_size;
-  switch (method_) {
-    case schema::ResizeMethod_BILINEAR: {
-      execute_format = schema::Format_NC4HW4;
-      output->SetFormat(schema::Format_NC4HW4);
-      exec_input_size = input->ElementsC4Num();
-      break;
-    }
-    case schema::ResizeMethod_NEAREST_NEIGHBOR: {
-      execute_format = schema::Format_NHWC;
-      output->SetFormat(schema::Format_NHWC);
-      exec_input_size = input->ElementsNum();
-      break;
-    }
-    default: {
-      MS_LOG(ERROR) << "Resize unknown method " << method_;
-      return RET_ERROR;
-    }
-  }
-
-  auto input_format = input->GetFormat();
-  if (input_format != execute_format) {
-    auto input_type = input->data_type();
-    layout_convertor_ = LayoutTransform(input_type, input_format, execute_format);
-    exec_input_data_ = reinterpret_cast<float *>(malloc(exec_input_size * sizeof(float)));
-    if (exec_input_data_ == nullptr) {
-      return RET_NULL_PTR;
-    }
+  if (!InferShapeDone()) {
+    return RET_OK;
   }
-
-  return RET_OK;
+  return ReSize();
 }
 
 int ResizeImpl(int task_id, LiteParallelGroupEnv *penv, void *cdata) {
   auto resize = reinterpret_cast<ResizeCPUKernel *>(cdata);
   auto error_code = resize->RunImpl(task_id);
-  if (error_code != NNACL_OK) {
+  if (error_code != RET_OK) {
     MS_LOG(ERROR) << "Resize Run error task_id[" << task_id << "] error_code[" << error_code << "]";
     return RET_ERROR;
   }
@@ -162,47 +59,30 @@ int ResizeCPUKernel::RunImpl(int task_id) {
     return RET_NULL_PTR;
   }
   auto input_shape = input->shape();
-  if (input_shape.size() != kRank) {
-    return RET_ERROR;
-  }
   if (context_ == nullptr) {
     return RET_NULL_PTR;
   }
 
   int ret = 0;
   switch (method_) {
-    case schema::ResizeMethod_BILINEAR: {
-      if (layout_convertor_ != nullptr) {
-        layout_convertor_(input_data, exec_input_data_, input->Batch(), input->Height() * input->Width(),
-                          input->Channel());
-        ret = ResizeBilinear(exec_input_data_, output_data, inputs_[0]->shape().data(), outputs_[0]->shape().data(),
+    case static_cast<int>(schema::ResizeMethod_BILINEAR): {
+      ret = ResizeBilinear(input_data, output_data, input_shape.data(), outputs_[0]->shape().data(),
                              align_corners_, task_id, context_->thread_num_);
-      } else {
-        ret = ResizeBilinear(input_data, output_data, inputs_[0]->shape().data(), outputs_[0]->shape().data(),
-                             align_corners_, task_id, context_->thread_num_);
-      }
       break;
     }
-    case schema::ResizeMethod_NEAREST_NEIGHBOR: {
+    case static_cast<int>(schema::ResizeMethod_NEAREST_NEIGHBOR): {
       if (align_corners_) {
         MS_LOG(ERROR) << "ResizeNearestNeighbor not support align_corners.";
         return RET_ERROR;
       }
-      if (layout_convertor_ != nullptr) {
-        layout_convertor_(input_data, exec_input_data_, input->Batch(), input->Height() * input->Width(),
-                          input->Channel());
-        ret = ResizeNearestNeighbor(exec_input_data_, output_data, input_shape.data(), outputs_[0]->shape().data(),
-                                    task_id, context_->thread_num_);
-      } else {
-        ret = ResizeNearestNeighbor(input_data, output_data, input_shape.data(), outputs_[0]->shape().data(), task_id,
+      ret = ResizeNearestNeighbor(input_data, output_data, input_shape.data(), outputs_[0]->shape().data(), task_id,
                                     context_->thread_num_);
-      }
       break;
     }
     case schema::ResizeMethod_UNKNOW:
     default: {
       MS_LOG(ERROR) << "Resize unknown method " << method_;
-      ret = NNACL_ERR;
+      ret = RET_ERROR;
     }
   }
   return ret;
@@ -221,31 +101,4 @@ int ResizeCPUKernel::Run() {
   }
   return RET_OK;
 }
-
-kernel::LiteKernel *CpuResizeFp32KernelCreator(const std::vector<lite::tensor::Tensor *> &inputs,
-                                               const std::vector<lite::tensor::Tensor *> &outputs,
-                                               OpParameter *opParameter, const lite::Context *ctx,
-                                               const kernel::KernelKey &desc, const lite::Primitive *primitive) {
-  if (opParameter == nullptr) {
-    MS_LOG(ERROR) << "Input opParameter is nullptr!";
-    return nullptr;
-  }
-  MS_ASSERT(desc.type == schema::PrimitiveType_Resize);
-  auto *kernel = new (std::nothrow) ResizeCPUKernel(opParameter, inputs, outputs, ctx, primitive);
-  if (kernel == nullptr) {
-    MS_LOG(ERROR) << "new ResizeCPUKernel fail!";
-    return nullptr;
-  }
-  auto ret = kernel->Init();
-  if (ret != RET_OK) {
-    delete kernel;
-    MS_LOG(ERROR) << "Init kernel failed, name: " << opParameter->name_ << ", type: "
-                  << schema::EnumNamePrimitiveType(static_cast<schema::PrimitiveType>(opParameter->type_));
-    return nullptr;
-  }
-
-  return kernel;
-}
-
-REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_Resize, CpuResizeFp32KernelCreator)
 }  // namespace mindspore::kernel

mindspore/lite/src/runtime/kernel/arm/fp32/resize.h

@@ -18,47 +18,25 @@
 
 #include <vector>
 #include "src/lite_kernel.h"
-#include "src/runtime/kernel/arm/nnacl/resize.h"
-#include "src/runtime/kernel/arm/base/layout_transform.h"
+#include "src/runtime/kernel/arm/base/resize_base.h"
 
 using mindspore::schema::PrimitiveType_Resize;
 using mindspore::schema::ResizeMethod;
 
 namespace mindspore::kernel {
-class ResizeCPUKernel : public LiteKernel {
+class ResizeCPUKernel : public ResizeBaseCPUKernel {
  public:
   ResizeCPUKernel(OpParameter *parameter, const std::vector<lite::tensor::Tensor *> &inputs,
                   const std::vector<lite::tensor::Tensor *> &outputs, const lite::Context *ctx,
                   const lite::Primitive *primitive)
-      : LiteKernel(parameter, inputs, outputs, ctx, primitive), context_(ctx) {}
+      : ResizeBaseCPUKernel(parameter, inputs, outputs, ctx, primitive) {}
 
-  ~ResizeCPUKernel() {
-    if (exec_input_data_ != nullptr) {
-      free(exec_input_data_);
-      exec_input_data_ = nullptr;
-    }
-  }
+  ~ResizeCPUKernel() = default;
 
   int Init() override;
   int ReSize() override { return 0; };
   int Run() override;
   int RunImpl(int task_id);
-
- protected:
-  const lite::Context *context_;
-
- private:
-  int CheckParameters();
-  int CheckInputsOuputs();
-
- private:
-  ResizeMethod method_;
-  int64_t new_height_;
-  int64_t new_width_;
-  bool align_corners_;
-  bool preserve_aspect_ratio;
-  LayoutConvertor layout_convertor_ = nullptr;
-  float *exec_input_data_ = nullptr;
 };
 }  // namespace mindspore::kernel
 

mindspore/lite/src/runtime/kernel/arm/int8/resize_int8.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <vector>
+#include "schema/model_generated.h"
+#include "src/kernel_registry.h"
+#include "nnacl/int8/resize.h"
+#include "include/errorcode.h"
+#include "src/runtime/kernel/arm/int8/resize_int8.h"
+#include "src/runtime/runtime_api.h"
+
+using mindspore::kernel::KERNEL_ARCH::kCPU;
+using mindspore::lite::KernelRegistrar;
+using mindspore::lite::RET_ERROR;
+using mindspore::lite::RET_INVALID_OP_ATTR;
+using mindspore::lite::RET_NULL_PTR;
+using mindspore::lite::RET_OK;
+
+namespace mindspore::kernel {
+namespace {
+constexpr int kInputNum = 1;
+constexpr int kOutputNum = 1;
+constexpr size_t kRank = 4;
+}  // namespace
+
+int ResizeInt8CPUKernel::Init() {
+  auto ret = ResizeBaseCPUKernel::Init();
+  if (ret != RET_OK) {
+    return ret;
+  }
+  quant_in_ = new (std::nothrow) QuantArg;
+  MS_ASSERT(quant_in_);
+  quant_out_ = new (std::nothrow) QuantArg;
+  MS_ASSERT(quant_out_);
+  auto input = inputs_.at(0);
+  quant_in_->zp_ = input->GetQuantParams().front().zeroPoint;
+  quant_in_->scale_ = input->GetQuantParams().front().scale;
+  auto output = outputs_.at(0);
+  quant_out_->zp_ = output->GetQuantParams().front().zeroPoint;
+  quant_out_->scale_ = output->GetQuantParams().front().scale;
+
+  multiplier_ = new (std::nothrow) QuantMulArg;
+  MS_ASSERT(multiplier_);
+  QuantizeRoundParameter(quant_in_->scale_ / quant_out_->scale_, &multiplier_->multiplier_, &multiplier_->left_shift_,
+                         &multiplier_->right_shift_);
+  if (!InferShapeDone()) {
+    return RET_OK;
+  }
+  return ReSize();
+}
+
+int ResizeInt8Impl(int task_id, LiteParallelGroupEnv *penv, void *cdata) {
+  auto resize = reinterpret_cast<ResizeInt8CPUKernel *>(cdata);
+  auto error_code = resize->RunImpl(task_id);
+  if (error_code != RET_OK) {
+    MS_LOG(ERROR) << "Resize Run error task_id[" << task_id << "] error_code[" << error_code << "]";
+    return RET_ERROR;
+  }
+  return RET_OK;
+}
+
+int ResizeInt8CPUKernel::RunImpl(int task_id) {
+  auto input = inputs_.at(0);
+  auto input_data = reinterpret_cast<const int8_t *>(input->Data());
+  if (input_data == nullptr) {
+    return RET_NULL_PTR;
+  }
+  auto output_data = reinterpret_cast<int8_t *>(outputs_.at(0)->Data());
+  if (output_data == nullptr) {
+    return RET_NULL_PTR;
+  }
+  auto input_shape = input->shape();
+
+  if (context_ == nullptr) {
+    return RET_NULL_PTR;
+  }
+
+  int ret = 0;
+  switch (method_) {
+    case static_cast<int>(schema::ResizeMethod_BILINEAR): {
+      ret = ResizeBilinearInt8(input_data, output_data, input_shape.data(), outputs_[0]->shape().data(), align_corners_,
+                               quant_in_, quant_out_, multiplier_, task_id, context_->thread_num_);
+
+      break;
+    }
+    case static_cast<int>(schema::ResizeMethod_NEAREST_NEIGHBOR): {
+      bool same_zp = quant_in_->zp_ == quant_out_->zp_;
+      bool same_scale = abs(quant_out_->scale_ - quant_in_->scale_) < 1e-6;
+      if (same_zp && same_scale) {
+        ret = ResizeNearestNeighborInt8Simple(input_data, output_data, input_shape.data(), outputs_[0]->shape().data(),
+                                              align_corners_, task_id, context_->thread_num_);
+      } else {
+        ret =
+          ResizeNearestNeighborInt8(input_data, output_data, input_shape.data(), outputs_[0]->shape().data(),
+                                    align_corners_, multiplier_, quant_in_, quant_out_, task_id, context_->thread_num_);
+      }
+      break;
+    }
+    case schema::ResizeMethod_UNKNOW:
+    default: {
+      MS_LOG(ERROR) << "Resize unknown method " << method_;
+      ret = RET_ERROR;
+    }
+  }
+  return ret;
+}
+
+int ResizeInt8CPUKernel::Run() {
+  auto ret = Prepare();
+  if (ret != RET_OK) {
+    MS_LOG(ERROR) << "Prepare failed.";
+    return RET_ERROR;
+  }
+  int error_code = LiteBackendParallelLaunch(ResizeInt8Impl, this, context_->thread_num_);
+  if (error_code != RET_OK) {
+    MS_LOG(ERROR) << "Resize run error, error_code[" << error_code << "]";
+    return RET_ERROR;
+  }
+  return RET_OK;
+}
+}  // namespace mindspore::kernel

mindspore/lite/src/runtime/kernel/arm/int8/resize_int8.h

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_RESIZE_INT8_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_RESIZE_INT8_H_
+
+#include <vector>
+#include "src/lite_kernel.h"
+#include "src/runtime/kernel/arm/base/resize_base.h"
+#include "nnacl/quantization/quantize.h"
+
+using mindspore::schema::PrimitiveType_Resize;
+using mindspore::schema::ResizeMethod;
+
+namespace mindspore::kernel {
+class ResizeInt8CPUKernel : public ResizeBaseCPUKernel {
+ public:
+  ResizeInt8CPUKernel(OpParameter *parameter, const std::vector<lite::tensor::Tensor *> &inputs,
+                      const std::vector<lite::tensor::Tensor *> &outputs, const lite::Context *ctx,
+                      const lite::Primitive *primitive)
+      : ResizeBaseCPUKernel(parameter, inputs, outputs, ctx, primitive) {}
+
+  ~ResizeInt8CPUKernel() {
+    delete quant_out_;
+    quant_out_ = nullptr;
+    delete quant_in_;
+    quant_in_ = nullptr;
+    delete multiplier_;
+    multiplier_ = nullptr;
+  }
+
+  int Init() override;
+  int ReSize() override { return 0; };
+  int Run() override;
+  int RunImpl(int task_id);
+
+ private:
+  QuantArg *quant_in_;
+  QuantArg *quant_out_;
+  QuantMulArg *multiplier_;
+};
+}  // namespace mindspore::kernel
+
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_RESIZE_INT8_H_

mindspore/lite/src/runtime/kernel/arm/nnacl/resize.cc → mindspore/lite/src/runtime/kernel/arm/nnacl/fp32/resize.cc

@@ -14,20 +14,20 @@
  * limitations under the License.
  */
 #include <math.h>
-#include "nnacl/resize.h"
+#include "nnacl/fp32/resize.h"
 #include "nnacl/common_func.h"
+#include "nnacl/errorcode.h"
 
 int ResizeBilinear(const float *input_data, float *output_data, const int *input_shape, const int *output_shape,
                    bool align_corners, int tid, int thread_num) {
   if (input_data == nullptr || output_data == nullptr || input_shape == nullptr || output_shape == nullptr) {
     return NNACL_NULL_PTR;
   }
-  // nhwc (memory layout is nc4hw4)
-  int n = input_shape[0];
+
+  int in_n = input_shape[0];
   int in_h = input_shape[1];
   int in_w = input_shape[2];
-  int channel = input_shape[3];
-  int c4 = UP_DIV(channel, C4NUM);
+  int in_c = input_shape[3];
 
   int new_height = output_shape[1];
   int new_width = output_shape[2];
@@ -40,67 +40,61 @@ int ResizeBilinear(const float *input_data, float *output_data, const int *input
     width_scale = (float)(in_w - 1) / (new_width - 1);
   }
 
-  int o[5];  // n c4 h w 4
-  for (o[0] = 0; o[0] < n; o[0]++) {
-    for (o[1] = tid; o[1] < c4; o[1] += thread_num) {
-      for (o[2] = 0; o[2] < new_height; o[2]++) {
-        float actual_y = (float)(o[2]) * height_scale;
-        int y_left = (int)(floor(actual_y));
-        int y_right = y_left + 1 < in_h ? (y_left + 1) : (in_h - 1);
-        float y_right_weight = actual_y - (float)(y_left);
-        float y_left_weight = 1.0 - y_right_weight;
-        for (o[3] = 0; o[3] < new_width; o[3]++) {
-          float actual_x = (float)(o[3]) * width_scale;
-          int x_left = (int)(floor(actual_x));
-          int x_right = x_left + 1 < in_w ? (x_left + 1) : (in_w - 1);
-          float x_right_weight = actual_x - (float)(x_left);
-          float x_left_weight = 1.0 - x_right_weight;
-
-          auto input_base_offset = (((o[0] * c4 + o[1]) * in_h + y_left) * in_w + x_left) * C4NUM;
-          auto output_base_offset = (((o[0] * c4 + o[1]) * new_height + o[2]) * new_width + o[3]) * C4NUM;
-          int in_offset_1_0 = (y_right - y_left) * in_w * C4NUM;
-          int in_offset_0_1 = (x_right - x_left) * C4NUM;
+  int n, h, w, c;
+  for (n = 0; n < in_n; n++) {
+    for (h = tid; h < new_height; h += thread_num) {
+      float actual_y = (float)h * height_scale;
+      int y_bottom = (int)(floor(actual_y));
+      int y_top = y_bottom + 1 < in_h ? (y_bottom + 1) : (in_h - 1);
+      float y_top_weight = actual_y - (float)(y_bottom);
+      float y_bottom_weight = 1.0f - y_top_weight;
+      for (w = 0; w < new_width; w++) {
+        float actual_x = (float)(w)*width_scale;
+        int x_left = (int)(floor(actual_x));
+        int x_right = x_left + 1 < in_w ? (x_left + 1) : (in_w - 1);
+        float x_right_weight = actual_x - (float)(x_left);
+        float x_left_weight = 1.0f - x_right_weight;
+        c = 0;
 #ifdef ENABLE_NEON
-          float32x4_t x_l_weight = vdupq_n_f32(x_left_weight);
-          float32x4_t x_r_weight = vdupq_n_f32(x_right_weight);
-          float32x4_t y_l_weight = vdupq_n_f32(y_left_weight);
-          float32x4_t y_r_weight = vdupq_n_f32(y_right_weight);
+        for (; c <= in_c - 4; c += 4) {
+          float32x4_t bottom_left = vld1q_f32(input_data + offset(input_shape, n, y_bottom, x_left, c));
+          float32x4_t bottom_right = vld1q_f32(input_data + offset(input_shape, n, y_bottom, x_right, c));
+          float32x4_t top_left = vld1q_f32(input_data + offset(input_shape, n, y_top, x_left, c));
+          float32x4_t top_right = vld1q_f32(input_data + offset(input_shape, n, y_top, x_right, c));
 
-          float32x4_t input_yl_xl = vld1q_f32(input_data + input_base_offset);
-          float32x4_t input_yr_xl = vld1q_f32(input_data + input_base_offset + in_offset_1_0);
-          float32x4_t input_yl_xr = vld1q_f32(input_data + input_base_offset + in_offset_0_1);
-          float32x4_t input_yr_xr = vld1q_f32(input_data + input_base_offset + in_offset_0_1 + in_offset_1_0);
+          float32x4_t y_top_w = vdupq_n_f32(y_top_weight);
+          float32x4_t y_bottom_w = vdupq_n_f32(y_bottom_weight);
+          float32x4_t x_left_w = vdupq_n_f32(x_left_weight);
+          float32x4_t x_right_w = vdupq_n_f32(x_right_weight);
 
           float32x4_t interp_value = vdupq_n_f32(0.0);
-          float32x4_t interp_value_tmp = vmulq_f32(input_yl_xl, y_l_weight);
-          interp_value_tmp = vmulq_f32(interp_value_tmp, x_l_weight);
-          interp_value = vaddq_f32(interp_value, interp_value_tmp);
+          float32x4_t tmp = vmulq_f32(bottom_left, y_bottom_w);
+          tmp = vmulq_f32(tmp, x_left_w);
+          interp_value = vaddq_f32(interp_value, tmp);
 
-          interp_value_tmp = vmulq_f32(input_yr_xl, y_r_weight);
-          interp_value_tmp = vmulq_f32(interp_value_tmp, x_l_weight);
-          interp_value = vaddq_f32(interp_value, interp_value_tmp);
+          tmp = vmulq_f32(bottom_right, y_bottom_w);
+          tmp = vmulq_f32(tmp, x_right_w);
+          interp_value = vaddq_f32(interp_value, tmp);
 
-          interp_value_tmp = vmulq_f32(input_yl_xr, y_l_weight);
-          interp_value_tmp = vmulq_f32(interp_value_tmp, x_r_weight);
-          interp_value = vaddq_f32(interp_value, interp_value_tmp);
+          tmp = vmulq_f32(top_left, y_top_w);
+          tmp = vmulq_f32(tmp, x_left_w);
+          interp_value = vaddq_f32(interp_value, tmp);
 
-          interp_value_tmp = vmulq_f32(input_yr_xr, y_r_weight);
-          interp_value_tmp = vmulq_f32(interp_value_tmp, x_r_weight);
-          interp_value = vaddq_f32(interp_value, interp_value_tmp);
-          vst1q_f32(output_base_offset + output_data, interp_value);
-#else
-          // 4 continuous data in a group;
-          for (o[4] = 0; o[4] < C4NUM; o[4]++) {
-            auto in_offset = input_base_offset + o[4];
-            auto output_offset = output_base_offset + o[4];
-            float interp_value =
-              input_data[in_offset] * y_left_weight * x_left_weight +
-              input_data[in_offset + in_offset_1_0] * y_right_weight * x_left_weight +
-              input_data[in_offset + in_offset_0_1] * y_left_weight * x_right_weight +
-              input_data[in_offset + in_offset_0_1 + in_offset_1_0] * y_right_weight * x_right_weight;
-            output_data[output_offset] = interp_value;
-          }
+          tmp = vmulq_f32(top_right, y_top_w);
+          tmp = vmulq_f32(tmp, x_right_w);
+          interp_value = vaddq_f32(interp_value, tmp);
+          vst1q_f32(output_data + offset(output_shape, n, h, w, c), interp_value);
+        }
 #endif
+        for (; c < in_c; c++) {
+          auto bottom_left = input_data[offset(input_shape, n, y_bottom, x_left, c)];
+          auto bottom_right = input_data[offset(input_shape, n, y_bottom, x_right, c)];
+          auto top_left = input_data[offset(input_shape, n, y_top, x_left, c)];
+          auto top_right = input_data[offset(input_shape, n, y_top, x_right, c)];
+          float interp_value = bottom_left * y_bottom_weight * x_left_weight +
+                               bottom_right * y_bottom_weight * x_right_weight +
+                               top_left * y_top_weight * x_left_weight + top_right * y_top_weight * x_right_weight;
+          output_data[offset(output_shape, n, h, w, c)] = interp_value;
         }
       }
     }
@@ -118,10 +112,10 @@ int ResizeNearestNeighbor(const float *input_data, float *output_data, const int
 
   for (batch = 0; batch < output_shape[0]; batch++) {
     for (y = tid; y < output_shape[1]; y += thread_num) {
-      int actual_y = (int)(floor((float)(y) * height_scale));
+      int actual_y = (int)(floor((float)(y)*height_scale));
       int input_y = actual_y < input_shape[1] ? actual_y : input_shape[1] - 1;
       for (x = 0; x < output_shape[2]; x++) {
-        int actual_x = (int)(floor((float)(x) * width_scale));
+        int actual_x = (int)(floor((float)(x)*width_scale));
         int input_x = actual_x < input_shape[2] ? actual_x : input_shape[2] - 1;
         int in_offset = offset(input_shape, batch, input_y, input_x, 0);
         int out_offset = offset(output_shape, batch, y, x, 0);
@@ -132,4 +126,3 @@ int ResizeNearestNeighbor(const float *input_data, float *output_data, const int
 
   return NNACL_OK;
 }
-

mindspore/lite/src/runtime/kernel/arm/nnacl/resize.h → mindspore/lite/src/runtime/kernel/arm/nnacl/fp32/resize.h

@@ -13,8 +13,8 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_RESIZE_H_
-#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_RESIZE_H_
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_FP32_RESIZE_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_FP32_RESIZE_H_
 
 #ifdef ENABLE_NEON
 #include <arm_neon.h>
@@ -22,18 +22,6 @@
 #include <memory.h>
 #include "nnacl/op_base.h"
 #include "schema/ops_generated.h"
-#include "nnacl/errorcode.h"
-
-using mindspore::schema::ResizeMethod;
-
-typedef struct ResizeParameter {
-  OpParameter op_parameter_;
-  ResizeMethod method_;
-  int64_t new_height_;
-  int64_t new_width_;
-  bool align_corners_;
-  bool preserve_aspect_ratio_;
-} ResizeParameter;
 
 int ResizeBilinear(const float *input_data, float *output_data, const int *input_shape, const int *output_shape,
                    bool align_corners, int tid, int thread_num);
@@ -41,5 +29,5 @@ int ResizeBilinear(const float *input_data, float *output_data, const int *input
 int ResizeNearestNeighbor(const float *input_data, float *output_data, const int *input_shape, const int *output_shape,
                           int tid, int thread_num);
 
-#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_RESIZE_H_
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_FP32_RESIZE_H_
 

mindspore/lite/src/runtime/kernel/arm/nnacl/int8/resize.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <math.h>
+#include "nnacl/int8/resize.h"
+#include "nnacl/common_func.h"
+#include "nnacl/quantization/fixed_point.h"
+#include "nnacl/errorcode.h"
+
+
+int ResizeBilinearInt8(const int8_t *input_data, int8_t *output_data, const int *input_shape, const int *output_shape,
+                       const bool align_corners, QuantArg *quant_in, QuantArg *quant_out, const QuantMulArg *mul_arg,
+                       int tid, int thread_num) {
+  if (input_data == nullptr || output_data == nullptr || input_shape == nullptr || output_shape == nullptr) {
+    return NNACL_NULL_PTR;
+  }
+
+  int32_t in_n = input_shape[0];
+  int32_t in_h = input_shape[1];
+  int32_t in_w = input_shape[2];
+  int32_t in_c = input_shape[3];
+
+  int32_t new_height = output_shape[1];
+  int32_t new_width = output_shape[2];
+  int32_t height_scale, width_scale;
+  ComputeScale(in_h, new_height, align_corners, &height_scale);
+  ComputeScale(in_w, new_width, align_corners, &width_scale);
+
+  int n, h, w, c;
+  for (n = 0; n < in_n; n++) {
+    for (h = tid; h < new_height; h += thread_num) {
+      //      float actual_y = (float)h * height_scale;
+      int base_offset = 20;
+      int scaled_actual_y;
+      int bottom, top;
+      int scaled_bottom_weight, scaled_top_weight;
+      ComputeInterpolationArgs(h, height_scale, in_h, &scaled_actual_y, &bottom, &scaled_bottom_weight, &top,
+                               &scaled_top_weight);
+      for (w = 0; w < new_width; w++) {
+        int scaled_actual_x;
+        int left, right;
+        int scaled_left_weight, scaled_right_weight;
+        ComputeInterpolationArgs(w, width_scale, in_w, &scaled_actual_x, &left, &scaled_left_weight, &right,
+                                 &scaled_right_weight);
+        for (c = 0; c < in_c; c++) {
+          const int64_t bottom_left_value =
+            (int64_t)(input_data[offset(input_shape, n, bottom, left, c)] - quant_in->zp_) * scaled_bottom_weight *
+            scaled_left_weight;
+          const int64_t bottom_right_value =
+            (int64_t)(input_data[offset(input_shape, n, bottom, right, c)] - quant_in->zp_) * scaled_bottom_weight *
+            scaled_right_weight;
+          const int64_t top_left_value = (int64_t)(input_data[offset(input_shape, n, top, left, c)] - quant_in->zp_) *
+                                         scaled_top_weight * scaled_left_weight;
+          const int64_t top_right_value = (int64_t)(input_data[offset(input_shape, n, top, right, c)] - quant_in->zp_) *
+                                          scaled_top_weight * scaled_right_weight;
+          const int64_t scaled_interp_value = bottom_left_value + bottom_right_value + top_left_value + top_right_value;
+          int32_t interp_value;
+          if (scaled_interp_value >= 0) {
+            interp_value = (scaled_interp_value + (1 << 19)) / (1 << 20);
+          } else {
+            interp_value = (scaled_interp_value - (1 << 19)) / (1 << 20);
+          }
+
+          const int out_interp_value =
+            MultiplyByQuantizedMultiplier(interp_value, mul_arg->multiplier_, mul_arg->left_shift_ + base_offset,
+                                          mul_arg->right_shift_ - base_offset) +
+            quant_out->zp_;
+          int8_t out_value;
+          out_value = out_interp_value > INT8_MAX ? INT8_MAX : out_interp_value;
+          out_value = out_value < INT8_MIN ? INT8_MIN : out_value;
+          output_data[offset(output_shape, n, h, w, c)] = out_value;
+        }
+      }
+    }
+  }
+  return NNACL_OK;
+}
+
+int ResizeNearestNeighborInt8Simple(const int8_t *input_data, int8_t *output_data, const int *input_shape,
+                                    const int *output_shape, const bool align_corners, int tid, int thread_num) {
+  int batch, y, x, c;
+  c = output_shape[3];
+  int in_h, in_w, new_height, new_width;
+  in_h = input_shape[1];
+  in_w = input_shape[2];
+  new_height = output_shape[1];
+  new_width = output_shape[2];
+
+  for (batch = 0; batch < output_shape[0]; batch++) {
+    for (y = tid; y < output_shape[1]; y += thread_num) {
+      int input_y;
+      ComputeNearestNeighborInt(y, in_h, new_height, align_corners, &input_y);
+      for (x = 0; x < output_shape[2]; x++) {
+        int input_x;
+        ComputeNearestNeighborInt(x, in_w, new_width, align_corners, &input_x);
+        int in_offset = offset(input_shape, batch, input_y, input_x, 0);
+        int out_offset = offset(output_shape, batch, y, x, 0);
+        memcpy(output_data + out_offset, input_data + in_offset, c * sizeof(int8_t));
+      }
+    }
+  }
+
+  return NNACL_OK;
+}
+
+void ComputeScale(const int32_t in_value, const int32_t out_value, const bool align_corners, int32_t *scale) {
+  *scale = (in_value * (1 << 10) + out_value / 2) / out_value;
+  if (align_corners && out_value > 1) {
+    *scale = ((in_value - 1) * (1 << 10) + (out_value - 1) / 2) / (out_value - 1);
+  }
+}
+
+void ComputeInterpolationArgs(const int32_t pos, const int32_t scale, const int32_t size, int32_t *scaled_pos,
+                              int32_t *low, int32_t *scaled_low_weight, int32_t *high, int32_t *scaled_high_weight) {
+  *scaled_pos = pos * scale;
+  int scale_back = *scaled_pos / (1 << 10);
+  *low = scale_back > 0 ? scale_back : 0;
+  *scaled_low_weight = (1 << 10) - (*scaled_pos - (1 << 10) * (*low));
+  *high = scale_back + 1 < size ? scale_back + 1 : size - 1;
+  *scaled_high_weight = *scaled_pos - (1 << 10) * (*low);
+}
+
+void ComputeNearestNeighborInt(const int32_t pos, const int in_size, const int32_t new_size, const bool align_corners,
+                               int32_t *nearest) {
+  *nearest = (in_size * pos) / new_size;
+  if (align_corners) {
+    *nearest = ((in_size - 1) * pos + (new_size - 1) / 2) / (new_size - 1);
+  }
+  *nearest = *nearest < in_size ? *nearest : in_size - 1;
+}
+
+int ResizeNearestNeighborInt8(const int8_t *input_data, int8_t *output_data, const int *input_shape,
+                              const int *output_shape, const bool align_corners, const QuantMulArg *multiplier,
+                              QuantArg *quant_in, QuantArg *quant_out, int tid, int thread_num) {
+  int base_offset = 20;
+  int32_t batch, y, x, c;
+  int32_t in_h, in_w, new_height, new_width;
+  in_h = input_shape[1];
+  in_w = input_shape[2];
+  new_height = output_shape[1];
+  new_width = output_shape[2];
+
+  for (batch = 0; batch < output_shape[0]; batch++) {
+    for (y = tid; y < output_shape[1]; y += thread_num) {
+      int input_y;
+      ComputeNearestNeighborInt(y, in_h, new_height, align_corners, &input_y);
+      for (x = 0; x < output_shape[2]; x++) {
+        int input_x;
+        ComputeNearestNeighborInt(x, in_w, new_width, align_corners, &input_x);
+        for (c = 0; c < output_shape[3]; c++) {
+          int in_offset = offset(input_shape, batch, input_y, input_x, c);
+          int out_offset = offset(output_shape, batch, y, x, c);
+
+          int32_t out_value = MultiplyByQuantizedMultiplier(
+                                input_data[in_offset] - quant_in->zp_, multiplier->multiplier_,
+                                multiplier->left_shift_ + base_offset, multiplier->right_shift_ - base_offset) +
+                              quant_out->zp_;
+          out_value = out_value > INT8_MAX ? INT8_MAX : out_value;
+          out_value = out_value < INT8_MIN ? INT8_MIN : out_value;
+          output_data[out_offset] = (int8_t)out_value;
+        }
+      }
+    }
+  }
+
+  return NNACL_OK;
+}

mindspore/lite/src/runtime/kernel/arm/nnacl/int8/resize.h

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_INT8_RESIZE_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_INT8_RESIZE_H_
+
+#ifdef ENABLE_NEON
+#include <arm_neon.h>
+#endif
+#include <memory.h>
+#include "nnacl/op_base.h"
+#include "nnacl/quantization/quantize.h"
+#include "nnacl/resize_parameter.h"
+
+int ResizeBilinearInt8(const int8_t *input_data, int8_t *output_data, const int *input_shape, const int *output_shape,
+                       const bool align_corners, QuantArg *quant_in, QuantArg *quant_out, const QuantMulArg *mul_arg,
+                       int tid, int thread_num);
+
+int ResizeNearestNeighborInt8Simple(const int8_t *input_data, int8_t *output_data, const int *input_shape,
+                                    const int *output_shape, const bool align_corners, int tid, int thread_num);
+
+int ResizeNearestNeighborInt8(const int8_t *input_data, int8_t *output_data, const int *input_shape,
+                              const int *output_shape, const bool align_corners, const QuantMulArg *multiplier,
+                              QuantArg *quant_in, QuantArg *quant_out, int tid, int thread_num);
+
+void ComputeScale(const int32_t in_value, const int32_t out_value, const bool align_corners, int32_t *scale);
+
+void ComputeInterpolationArgs(const int32_t pos, const int32_t scale, const int32_t size, int32_t *scaled_pos,
+                              int32_t *low, int32_t *scaled_low_weight, int32_t *high, int32_t *scaled_high_weight);
+
+void ComputeNearestNeighborInt(const int32_t pos, const int in_size, const int32_t new_size, const bool align_corners,
+                               int32_t *nearest);
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_INT8_RESIZE_H_

mindspore/lite/src/runtime/kernel/arm/nnacl/quantization/quantize.h

@@ -167,6 +167,12 @@ typedef struct ReshapeQuantArg {
   int output_activation_max_;
 } ReshapeQuantArg;
 
+typedef struct QuantMulArg {
+  int32_t multiplier_;
+  int left_shift_;
+  int right_shift_;
+} QuantMulArg;
+
 void QuantizeMultiplier(double double_multiplier, int32_t *quantized_multiplier, int *shift);
 
 inline void QuantizeMultiplierSmallerThanOne(double double_multiplier, int32_t *quantized_multiplier,

mindspore/lite/src/runtime/kernel/arm/nnacl/resize_parameter.h

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_RESIZE_PARAMETER_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_RESIZE_PARAMETER_H_
+
+#include "src/runtime/kernel/arm/nnacl/op_base.h"
+typedef struct ResizeParameter {
+  OpParameter op_parameter_;
+  int method_;
+  int64_t new_height_;
+  int64_t new_width_;
+  bool align_corners_;
+  bool preserve_aspect_ratio_;
+} ResizeParameter;
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_NNACL_RESIZE_PARAMETER_H_

mindspore/lite/test/common/common_test.h

@@ -56,6 +56,19 @@ class Common : public testing::Test {
     }
   }
 
+  void CompareOutputInt8(int8_t *output_data, int8_t *correct_data, int size, float err_percent) {
+    int bias_count = 0;
+    for (size_t i = 0; i < size; i++) {
+      int8_t diff = abs(output_data[i] - correct_data[i]);
+      ASSERT_LE(diff, 1);
+      if (diff == 1) {
+        bias_count++;
+      }
+    }
+    float bias_percent = static_cast<float>(bias_count) / size;
+    ASSERT_LE(bias_percent, err_percent);
+  }
+
   void ReadFile(const char *file, size_t *size, char **buf) {
     ASSERT_NE(nullptr, file);
     ASSERT_NE(nullptr, size);

mindspore/lite/test/ut/src/runtime/kernel/arm/fp32/reduce_fp32_tests.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <memory>
+#include "utils/log_adapter.h"
+#include "common/common_test.h"
+#include "mindspore/lite/src/runtime/kernel/arm/nnacl/fp32/reduce.h"
+
+namespace mindspore {
+
+class TestReduceFp32 : public mindspore::Common {
+ public:
+  TestReduceFp32() = default;
+  int tid = 0;
+  int thread_num = 1;
+  float err_tol = 1e-5;
+};
+
+TEST_F(TestReduceFp32, Mean) {
+  /* 2 4 4 3 NHWC */
+  float in[96] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
+                  16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0,
+                  32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                  48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0,
+                  64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0,
+                  80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+  float correct[24] = {18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0,
+                       66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0};
+
+  int input_shape[4] = {2, 4, 4, 3};
+  // int output_shape[4] = {2, 1, 4, 3};
+
+  float out[24] = {0};
+  int outer_size = 2;
+  int inner_size = 12;
+  int axis_size = 4;
+  (void)ReduceMean(outer_size, inner_size, axis_size, in, input_shape, out, tid, thread_num);
+
+  int output_size = 24;
+  CompareOutputData(out, correct, output_size, err_tol);
+}
+
+TEST_F(TestReduceFp32, Mean2Thread) {
+  /* 2*4*4*3 NHWC */
+  float in[96] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
+                  16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0,
+                  32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                  48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0,
+                  64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0,
+                  80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+  float correct[24] = {18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0,
+                       66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0};
+
+  int input_shape[4] = {2, 4, 4, 3};
+  // int output_shape[4] = {2, 1, 4, 3};
+
+  float out[24] = {0};
+  int outer_size = 2;
+  int inner_size = 12;
+  int axis_size = 4;
+  thread_num = 2;
+  tid = 0;
+  (void)ReduceMean(outer_size, inner_size, axis_size, in, input_shape, out, tid, thread_num);
+  tid = 1;
+  (void)ReduceMean(outer_size, inner_size, axis_size, in, input_shape, out, tid, thread_num);
+
+  int output_size = 24;
+  CompareOutputData(out, correct, output_size, err_tol);
+}
+
+TEST_F(TestReduceFp32, MeanAllAxis) {
+  /* 2*4*4*3 NHWC */
+  float in[96] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
+                  16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0,
+                  32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                  48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0,
+                  64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0,
+                  80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+  float correct[1] = {47.5};
+  float out[1] = {0};
+
+  int input_shape[4] = {2, 4, 4, 3};
+  int outer_size = 1;
+  int inner_size = 48;
+  int axis_size = 2;
+  float *src = in;
+  float dst1[48] = {0};
+  MS_ASSERT(dst != nullptr);
+  (void)ReduceMean(outer_size, inner_size, axis_size, src, input_shape, dst1, tid, thread_num);
+
+  input_shape[0] = 1;  // 1 4 4 3
+  outer_size = 1;
+  inner_size = 12;
+  axis_size = 4;
+  src = dst1;
+  float dst2[12] = {0};
+  (void)ReduceMean(outer_size, inner_size, axis_size, src, input_shape, dst2, tid, thread_num);
+
+  input_shape[1] = 1;  // 1 1 4 3
+  outer_size = 1;
+  inner_size = 3;
+  axis_size = 4;
+  src = dst2;
+  float dst3[3] = {0};
+  (void)ReduceMean(outer_size, inner_size, axis_size, src, input_shape, dst3, tid, thread_num);
+
+  input_shape[2] = 1;  // 1 1 1 3
+  outer_size = 1;
+  inner_size = 1;
+  axis_size = 3;
+  src = dst3;
+  (void)ReduceMean(outer_size, inner_size, axis_size, src, input_shape, out, tid, thread_num);
+
+  int output_size = 1;
+  CompareOutputData(out, correct, output_size, err_tol);
+}
+
+TEST_F(TestReduceFp32, Sum) {
+  /* 2*4*4*3 NHWC */
+  float in[96] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
+                  16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0,
+                  32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                  48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0,
+                  64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0,
+                  80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+  float correct[24] = {72.0,  76.0,  80.0,  84.0,  88.0,  92.0,  96.0,  100.0, 104.0, 108.0, 112.0, 116.0,
+                       264.0, 268.0, 272.0, 276.0, 280.0, 284.0, 288.0, 292.0, 296.0, 300.0, 304.0, 308.0};
+
+  int input_shape[4] = {2, 4, 4, 3};
+  // int output_shape[4] = {2, 1, 4, 3};
+
+  float out[24] = {0};
+  int outer_size = 2;
+  int inner_size = 12;
+  int axis_size = 4;
+  (void)ReduceSum(outer_size, inner_size, axis_size, in, input_shape, out, tid, thread_num);
+
+  int output_size = 24;
+  CompareOutputData(out, correct, output_size, err_tol);
+}
+
+TEST_F(TestReduceFp32, Sum2Thread) {
+  /* 2*4*4*3 NHWC */
+  float in[96] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
+                  16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0,
+                  32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                  48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0,
+                  64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0,
+                  80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+  float correct[24] = {72.0,  76.0,  80.0,  84.0,  88.0,  92.0,  96.0,  100.0, 104.0, 108.0, 112.0, 116.0,
+                       264.0, 268.0, 272.0, 276.0, 280.0, 284.0, 288.0, 292.0, 296.0, 300.0, 304.0, 308.0};
+
+  int input_shape[4] = {2, 4, 4, 3};
+  // int output_shape[4] = {2, 1, 4, 3};
+
+  float out[24] = {0};
+  int outer_size = 2;
+  int inner_size = 12;
+  int axis_size = 4;
+  thread_num = 2;
+  tid = 0;
+  (void)ReduceSum(outer_size, inner_size, axis_size, in, input_shape, out, tid, thread_num);
+  tid = 1;
+  (void)ReduceSum(outer_size, inner_size, axis_size, in, input_shape, out, tid, thread_num);
+
+  int output_size = 24;
+  CompareOutputData(out, correct, output_size, err_tol);
+}
+
+TEST_F(TestReduceFp32, SumAllAxis) {
+  /* 2*4*4*3 NHWC */
+  float in[96] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
+                  16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0,
+                  32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                  48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0,
+                  64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0,
+                  80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+  float correct[1] = {4560};
+  float out[1] = {0};
+
+  int input_shape[4] = {2, 4, 4, 3};
+  int outer_size = 1;
+  int inner_size = 48;
+  int axis_size = 2;
+  float *src = in;
+  float dst1[48] = {0};
+  MS_ASSERT(dst != nullptr);
+  (void)ReduceSum(outer_size, inner_size, axis_size, src, input_shape, dst1, tid, thread_num);
+
+  input_shape[0] = 1;  // 1 4 4 3
+  outer_size = 1;
+  inner_size = 12;
+  axis_size = 4;
+  src = dst1;
+  float dst2[12] = {0};
+  (void)ReduceSum(outer_size, inner_size, axis_size, src, input_shape, dst2, tid, thread_num);
+
+  input_shape[1] = 1;  // 1 1 4 3
+  outer_size = 1;
+  inner_size = 3;
+  axis_size = 4;
+  src = dst2;
+  float dst3[3] = {0};
+  (void)ReduceSum(outer_size, inner_size, axis_size, src, input_shape, dst3, tid, thread_num);
+
+  input_shape[2] = 1;  // 1 1 1 3
+  outer_size = 1;
+  inner_size = 1;
+  axis_size = 3;
+  src = dst3;
+  (void)ReduceSum(outer_size, inner_size, axis_size, src, input_shape, out, tid, thread_num);
+
+  int output_size = 1;
+  CompareOutputData(out, correct, output_size, err_tol);
+}
+
+TEST_F(TestReduceFp32, Max) {
+  /* 2*4*4*3 NHWC */
+  float in[96] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
+                  16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0,
+                  32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                  48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0,
+                  64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0,
+                  80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+  float correct[24] = {36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                       84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+
+  int input_shape[4] = {2, 4, 4, 3};
+  // int output_shape[4] = {2, 1, 4, 3};
+
+  float out[24] = {0};
+  int outer_size = 2;
+  int inner_size = 12;
+  int axis_size = 4;
+  (void)ReduceMax(outer_size, inner_size, axis_size, in, input_shape, out, tid, thread_num);
+
+  int output_size = 24;
+  CompareOutputData(out, correct, output_size, err_tol);
+}
+
+TEST_F(TestReduceFp32, Min) {
+  /* 2*4*4*3 NHWC */
+  float in[96] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
+                  16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0,
+                  32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                  48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0,
+                  64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0,
+                  80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+  float correct[24] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0,
+                       48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0};
+
+  int input_shape[4] = {2, 4, 4, 3};
+  // int output_shape[4] = {2, 1, 4, 3};
+
+  float out[24] = {0};
+  int outer_size = 2;
+  int inner_size = 12;
+  int axis_size = 4;
+  (void)ReduceMin(outer_size, inner_size, axis_size, in, input_shape, out, tid, thread_num);
+
+  int output_size = 24;
+  CompareOutputData(out, correct, output_size, err_tol);
+}
+
+TEST_F(TestReduceFp32, Prod) {
+  /* 2*4*4*3 NHWC */
+  float in[96] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
+                  16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0,
+                  32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                  48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0,
+                  64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0,
+                  80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+  float correct[24] = {0.0,        12025.0,    27664.0,    47385.0,    71680.0,    101065.0,   136080.0,   177289.0,
+                       225280.0,   280665.0,   344080.0,   416185.0,   17418240.0, 18546744.0, 19728400.0, 20964824.0,
+                       22257664.0, 23608584.0, 25019280.0, 26491464.0, 28026880.0, 29627288.0, 31294480.0, 33030264.0};
+
+  int input_shape[4] = {2, 4, 4, 3};
+  // int output_shape[4] = {2, 1, 4, 3};
+
+  float out[24] = {0};
+  int outer_size = 2;
+  int inner_size = 12;
+  int axis_size = 4;
+  (void)ReduceProd(outer_size, inner_size, axis_size, in, input_shape, out, tid, thread_num);
+
+  int output_size = 24;
+  CompareOutputData(out, correct, output_size, err_tol);
+}
+
+TEST_F(TestReduceFp32, SumSquare) {
+  /* 2*4*4*3 NHWC */
+  float in[96] = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
+                  16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0,
+                  32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 40.0, 41.0, 42.0, 43.0, 44.0, 45.0, 46.0, 47.0,
+                  48.0, 49.0, 50.0, 51.0, 52.0, 53.0, 54.0, 55.0, 56.0, 57.0, 58.0, 59.0, 60.0, 61.0, 62.0, 63.0,
+                  64.0, 65.0, 66.0, 67.0, 68.0, 69.0, 70.0, 71.0, 72.0, 73.0, 74.0, 75.0, 76.0, 77.0, 78.0, 79.0,
+                  80.0, 81.0, 82.0, 83.0, 84.0, 85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0, 95.0};
+  float correct[24] = {2016.0,  2164.0,  2320.0,  2484.0,  2656.0,  2836.0,  3024.0,  3220.0,
+                       3424.0,  3636.0,  3856.0,  4084.0,  18144.0, 18676.0, 19216.0, 19764.0,
+                       20320.0, 20884.0, 21456.0, 22036.0, 22624.0, 23220.0, 23824.0, 24436.0};
+
+  int input_shape[4] = {2, 4, 4, 3};
+  // int output_shape[4] = {2, 1, 4, 3};
+
+  float out[24] = {0};
+  int outer_size = 2;
+  int inner_size = 12;
+  int axis_size = 4;
+  (void)ReduceSumSquare(outer_size, inner_size, axis_size, in, input_shape, out, tid, thread_num);
+
+  int output_size = 24;
+  CompareOutputData(out, correct, output_size, err_tol);
+}
+}  // namespace mindspore

mindspore/lite/test/ut/src/runtime/kernel/arm/fp32/resize_bilinear_fp32_tests.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <iostream>
+#include <vector>
+#include "common/common_test.h"
+#include "mindspore/lite/src/runtime/kernel/arm/nnacl/fp32/resize.h"
+
+namespace mindspore {
+
+class TestResizeBilinearFp32 : public mindspore::Common {
+ public:
+  TestResizeBilinearFp32() = default;
+
+ public:
+  int tid = 0;
+  int thread_num = 1;
+  float err_tol = 1e-5;
+};
+
+// 1*1 -> 1*1
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest1) {
+  std::vector<float> input = {1.0};
+  std::vector<int> input_shape = {1, 1, 1, 1};
+  std::vector<int> output_shape = {1, 1, 1, 1};
+  std::vector<float> expect = {1.0};
+  bool align_corners = false;
+
+  auto output_size = 1;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 1*1
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest2) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 1, 1, 1};
+  std::vector<float> expect = {0.0};
+  bool align_corners = false;
+
+  int output_size = 1;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 1*2
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest3) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 1, 2, 1};
+  std::vector<float> expect = {0.0, 1.0};
+  bool align_corners = false;
+
+  auto output_size = 2;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 2*1
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest4) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 2, 1, 1};
+  std::vector<float> expect = {0.0, 2.0};
+  bool align_corners = false;
+
+  auto output_size = 2;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 2*2
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest5) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 2, 2, 1};
+  std::vector<float> expect = {0.0, 1.0, 2.0, 3.0};
+  bool align_corners = false;
+
+  auto output_size = 4;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 1*4
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest6) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 1, 4, 1};
+  std::vector<float> expect = {0.0, 0.5, 1.0, 1.0};
+  bool align_corners = false;
+
+  auto output_size = 4;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 4*1
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest7) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 4, 1, 1};
+  std::vector<float> expect = {0.0, 1.0, 2.0, 2.0};
+  bool align_corners = false;
+
+  auto output_size = 4;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 2*4
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest8) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 2, 4, 1};
+  std::vector<float> expect = {0.0, 0.5, 1.0, 1.0, 2.0, 2.5, 3.0, 3.0};
+  bool align_corners = false;
+
+  auto output_size = 8;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 4*2
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest9) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 4, 2, 1};
+  std::vector<float> expect = {0.0, 1.0, 1.0, 2.0, 2.0, 3.0, 2.0, 3.0};
+  bool align_corners = false;
+
+  auto output_size = 8;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 3*3
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest10) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 3, 3, 1};
+  std::vector<float> expect = {0.0, 0.6666667, 1.0, 1.3333334, 2.0, 2.3333335, 2.0, 2.6666667, 3.0};
+  bool align_corners = false;
+
+  auto output_size = 9;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 4*4
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest11) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 4, 4, 1};
+  std::vector<float> expect = {0.0, 0.5, 1.0, 1.0, 1.0, 1.5, 2.0, 2.0, 2.0, 2.5, 3.0, 3.0, 2.0, 2.5, 3.0, 3.0};
+  bool align_corners = false;
+
+  auto output_size = 16;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2*2*5 -> 2*4*4*5
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest12) {
+  std::vector<float> input = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0,
+                              14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0,
+                              28.0, 29.0, 30.0, 31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  std::vector<int> input_shape = {2, 2, 2, 5};
+  std::vector<int> output_shape = {2, 4, 4, 5};
+  std::vector<float> expect = {
+    0.0,  1.0,  2.0,  3.0,  4.0,  2.5,  3.5,  4.5,  5.5,  6.5,  5.0,  6.0,  7.0,  8.0,  9.0,  5.0,  6.0,  7.0,
+    8.0,  9.0,  5.0,  6.0,  7.0,  8.0,  9.0,  7.5,  8.5,  9.5,  10.5, 11.5, 10.0, 11.0, 12.0, 13.0, 14.0, 10.0,
+    11.0, 12.0, 13.0, 14.0, 10.0, 11.0, 12.0, 13.0, 14.0, 12.5, 13.5, 14.5, 15.5, 16.5, 15.0, 16.0, 17.0, 18.0,
+    19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 10.0, 11.0, 12.0, 13.0, 14.0, 12.5, 13.5, 14.5, 15.5, 16.5, 15.0, 16.0,
+    17.0, 18.0, 19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 22.5, 23.5, 24.5, 25.5, 26.5,
+    25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 27.5, 28.5, 29.5,
+    30.5, 31.5, 30.0, 31.0, 32.0, 33.0, 34.0, 30.0, 31.0, 32.0, 33.0, 34.0, 30.0, 31.0, 32.0, 33.0, 34.0, 32.5,
+    33.5, 34.5, 35.5, 36.5, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0, 30.0, 31.0, 32.0, 33.0,
+    34.0, 32.5, 33.5, 34.5, 35.5, 36.5, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  bool align_corners = false;
+
+  auto output_size = 160;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2*2*5 -> 2*4*4*5 align corners
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest13) {
+  std::vector<float> input = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0,
+                              14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0,
+                              28.0, 29.0, 30.0, 31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  std::vector<int> input_shape = {2, 2, 2, 5};
+  std::vector<int> output_shape = {2, 4, 4, 5};
+  std::vector<float> expect = {
+    0.0,       1.0,       2.0,       3.0,       4.0,       1.6666667, 2.6666667, 3.6666667, 4.666667,  5.666667,
+    3.3333335, 4.3333335, 5.3333335, 6.3333335, 7.3333335, 5.0,       6.0,       7.0,       8.0,       9.0,
+    3.3333335, 4.3333335, 5.3333335, 6.3333335, 7.3333335, 5.0,       6.0,       7.0,       8.0,       9.0,
+    6.666667,  7.666667,  8.666667,  9.666667,  10.666667, 8.333334,  9.333334,  10.333334, 11.333334, 12.333334,
+    6.666667,  7.666667,  8.666667,  9.666667,  10.666667, 8.333334,  9.333334,  10.333334, 11.333334, 12.333334,
+    10.0,      11.0,      12.0,      13.0,      14.0,      11.666667, 12.666667, 13.666667, 14.666667, 15.666667,
+    10.0,      11.0,      12.0,      13.0,      14.0,      11.666667, 12.666667, 13.666667, 14.666667, 15.666667,
+    13.333334, 14.333334, 15.333334, 16.333334, 17.333334, 15.0,      16.0,      17.0,      18.0,      19.0,
+    20.0,      21.0,      22.0,      23.0,      24.0,      21.666666, 22.666666, 23.666666, 24.666666, 25.666666,
+    23.333334, 24.333334, 25.333334, 26.333334, 27.333334, 25.0,      26.0,      27.0,      28.0,      29.0,
+    23.333334, 24.333334, 25.333334, 26.333334, 27.333334, 25.0,      26.0,      27.0,      28.0,      29.0,
+    26.666666, 27.666666, 28.666666, 29.666666, 30.666666, 28.333334, 29.333334, 30.333334, 31.333334, 32.333332,
+    26.666668, 27.666668, 28.666668, 29.666668, 30.666668, 28.333332, 29.333334, 30.333334, 31.333334, 32.333336,
+    30.0,      31.0,      32.0,      33.0,      34.0,      31.666668, 32.666668, 33.666668, 34.666668, 35.666668,
+    30.0,      31.0,      32.0,      33.0,      34.0,      31.666666, 32.666668, 33.666668, 34.666668, 35.666668,
+    33.333332, 34.333332, 35.333332, 36.333332, 37.333332, 35.0,      36.0,      37.0,      38.0,      39.0};
+  bool align_corners = true;
+
+  auto output_size = 160;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2*2*5 -> 2*4*4*5 thread_num 2
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest14) {
+  std::vector<float> input = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0,
+                              14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0,
+                              28.0, 29.0, 30.0, 31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  std::vector<int> input_shape = {2, 2, 2, 5};
+  std::vector<int> output_shape = {2, 4, 4, 5};
+  std::vector<float> expect = {
+    0.0,  1.0,  2.0,  3.0,  4.0,  2.5,  3.5,  4.5,  5.5,  6.5,  5.0,  6.0,  7.0,  8.0,  9.0,  5.0,  6.0,  7.0,
+    8.0,  9.0,  5.0,  6.0,  7.0,  8.0,  9.0,  7.5,  8.5,  9.5,  10.5, 11.5, 10.0, 11.0, 12.0, 13.0, 14.0, 10.0,
+    11.0, 12.0, 13.0, 14.0, 10.0, 11.0, 12.0, 13.0, 14.0, 12.5, 13.5, 14.5, 15.5, 16.5, 15.0, 16.0, 17.0, 18.0,
+    19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 10.0, 11.0, 12.0, 13.0, 14.0, 12.5, 13.5, 14.5, 15.5, 16.5, 15.0, 16.0,
+    17.0, 18.0, 19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 22.5, 23.5, 24.5, 25.5, 26.5,
+    25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 27.5, 28.5, 29.5,
+    30.5, 31.5, 30.0, 31.0, 32.0, 33.0, 34.0, 30.0, 31.0, 32.0, 33.0, 34.0, 30.0, 31.0, 32.0, 33.0, 34.0, 32.5,
+    33.5, 34.5, 35.5, 36.5, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0, 30.0, 31.0, 32.0, 33.0,
+    34.0, 32.5, 33.5, 34.5, 35.5, 36.5, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  bool align_corners = false;
+
+  auto output_size = 160;
+  std::vector<float> output(output_size, 0.0);
+  thread_num = 2;
+  tid = 0;
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  tid = 1;
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2*2*5 -> 2*4*4*5 thread_num 4
+TEST_F(TestResizeBilinearFp32, ResizeBilinearTest15) {
+  std::vector<float> input = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0,
+                              14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0,
+                              28.0, 29.0, 30.0, 31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  std::vector<int> input_shape = {2, 2, 2, 5};
+  std::vector<int> output_shape = {2, 4, 4, 5};
+  std::vector<float> expect = {
+    0.0,  1.0,  2.0,  3.0,  4.0,  2.5,  3.5,  4.5,  5.5,  6.5,  5.0,  6.0,  7.0,  8.0,  9.0,  5.0,  6.0,  7.0,
+    8.0,  9.0,  5.0,  6.0,  7.0,  8.0,  9.0,  7.5,  8.5,  9.5,  10.5, 11.5, 10.0, 11.0, 12.0, 13.0, 14.0, 10.0,
+    11.0, 12.0, 13.0, 14.0, 10.0, 11.0, 12.0, 13.0, 14.0, 12.5, 13.5, 14.5, 15.5, 16.5, 15.0, 16.0, 17.0, 18.0,
+    19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 10.0, 11.0, 12.0, 13.0, 14.0, 12.5, 13.5, 14.5, 15.5, 16.5, 15.0, 16.0,
+    17.0, 18.0, 19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 22.5, 23.5, 24.5, 25.5, 26.5,
+    25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 27.5, 28.5, 29.5,
+    30.5, 31.5, 30.0, 31.0, 32.0, 33.0, 34.0, 30.0, 31.0, 32.0, 33.0, 34.0, 30.0, 31.0, 32.0, 33.0, 34.0, 32.5,
+    33.5, 34.5, 35.5, 36.5, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0, 30.0, 31.0, 32.0, 33.0,
+    34.0, 32.5, 33.5, 34.5, 35.5, 36.5, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  bool align_corners = false;
+
+  auto output_size = 160;
+  std::vector<float> output(output_size, 0.0);
+  thread_num = 4;
+  tid = 0;
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  tid = 1;
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  tid = 2;
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  tid = 3;
+  ResizeBilinear(input.data(), output.data(), input_shape.data(), output_shape.data(), align_corners, tid,
+                 thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+}  // namespace mindspore

mindspore/lite/test/ut/src/runtime/kernel/arm/fp32/resize_nearest_neighbor_fp32_tests.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <vector>
+#include "common/common_test.h"
+#include "mindspore/lite/src/runtime/kernel/arm/nnacl/fp32/resize.h"
+
+namespace mindspore {
+
+class TestResizeNearestNeighborFp32 : public mindspore::Common {
+ public:
+  TestResizeNearestNeighborFp32() = default;
+
+ public:
+  int tid = 0;
+  int thread_num = 1;
+  float err_tol = 1e-5;
+};
+
+// 1*1 -> 1*1
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest1) {
+  std::vector<float> input = {1.0};
+  std::vector<int> input_shape = {1, 1, 1, 1};
+  std::vector<int> output_shape = {1, 1, 1, 1};
+  std::vector<float> expect = {1.0};
+  size_t output_size = 1;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 1*1
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest2) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 1, 1, 1};
+  std::vector<float> expect = {0.0};
+  size_t output_size = 1;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 1*2
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest3) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 1, 2, 1};
+  std::vector<float> expect = {0.0, 1.0};
+  size_t output_size = 2;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 2*1
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest4) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 2, 1, 1};
+  std::vector<float> expect = {0.0, 2.0};
+  size_t output_size = 2;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 2*2
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest5) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 2, 2, 1};
+  std::vector<float> expect = {0.0, 1.0, 2.0, 3.0};
+  size_t output_size = 4;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 1*4
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest6) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 1, 4, 1};
+  std::vector<float> expect = {0.0, 0.0, 1.0, 1.0};
+  size_t output_size = 4;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 4*1
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest7) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 4, 1, 1};
+  std::vector<float> expect = {0.0, 0.0, 2.0, 2.0};
+  size_t output_size = 4;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 2*4
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest8) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 2, 4, 1};
+  std::vector<float> expect = {0.0, 0.0, 1.0, 1.0, 2.0, 2.0, 3.0, 3.0};
+  size_t output_size = 8;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 4*2
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest9) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 4, 2, 1};
+  std::vector<float> expect = {0.0, 1.0, 0.0, 1.0, 2.0, 3.0, 2.0, 3.0};
+  size_t output_size = 8;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 3*3
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest10) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 3, 3, 1};
+  std::vector<float> expect = {0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 2.0, 2.0, 3.0};
+  size_t output_size = 9;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2 -> 4*4
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest11) {
+  std::vector<float> input = {0.0, 1.0, 2.0, 3.0};
+  std::vector<int> input_shape = {1, 2, 2, 1};
+  std::vector<int> output_shape = {1, 4, 4, 1};
+  std::vector<float> expect = {0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 2.0, 2.0, 3.0, 3.0, 2.0, 2.0, 3.0, 3.0};
+  size_t output_size = 16;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2*2*5 -> 2*4*4*5
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest12) {
+  std::vector<float> input = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0,
+                              14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0,
+                              28.0, 29.0, 30.0, 31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  std::vector<int> input_shape = {2, 2, 2, 5};
+  std::vector<int> output_shape = {2, 4, 4, 5};
+  std::vector<float> expect = {
+    0.0,  1.0,  2.0,  3.0,  4.0,  0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  5.0,  6.0,  7.0,
+    8.0,  9.0,  0.0,  1.0,  2.0,  3.0,  4.0,  0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  5.0,
+    6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0, 17.0, 18.0,
+    19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 10.0, 11.0, 12.0, 13.0, 14.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0,
+    17.0, 18.0, 19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 20.0, 21.0, 22.0, 23.0, 24.0,
+    25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 20.0, 21.0, 22.0, 23.0, 24.0, 20.0, 21.0, 22.0,
+    23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0, 32.0, 33.0, 34.0, 30.0,
+    31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0, 30.0, 31.0, 32.0, 33.0,
+    34.0, 30.0, 31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  size_t output_size = 160;
+  std::vector<float> output(output_size, 0.0);
+
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2*2*5 -> 2*4*4*5 thread_num 2
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest13) {
+  std::vector<float> input = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0,
+                              14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0,
+                              28.0, 29.0, 30.0, 31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  std::vector<int> input_shape = {2, 2, 2, 5};
+  std::vector<int> output_shape = {2, 4, 4, 5};
+  std::vector<float> expect = {
+    0.0,  1.0,  2.0,  3.0,  4.0,  0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  5.0,  6.0,  7.0,
+    8.0,  9.0,  0.0,  1.0,  2.0,  3.0,  4.0,  0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  5.0,
+    6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0, 17.0, 18.0,
+    19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 10.0, 11.0, 12.0, 13.0, 14.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0,
+    17.0, 18.0, 19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 20.0, 21.0, 22.0, 23.0, 24.0,
+    25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 20.0, 21.0, 22.0, 23.0, 24.0, 20.0, 21.0, 22.0,
+    23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0, 32.0, 33.0, 34.0, 30.0,
+    31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0, 30.0, 31.0, 32.0, 33.0,
+    34.0, 30.0, 31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  size_t output_size = 160;
+  std::vector<float> output(output_size, 0.0);
+
+  thread_num = 2;
+  tid = 0;
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  tid = 1;
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+
+// 2*2*2*5 -> 2*4*4*5 thread_num 4
+TEST_F(TestResizeNearestNeighborFp32, ResizeNearestNeighborTest14) {
+  std::vector<float> input = {0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0,
+                              14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0,
+                              28.0, 29.0, 30.0, 31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  std::vector<int> input_shape = {2, 2, 2, 5};
+  std::vector<int> output_shape = {2, 4, 4, 5};
+  std::vector<float> expect = {
+    0.0,  1.0,  2.0,  3.0,  4.0,  0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  5.0,  6.0,  7.0,
+    8.0,  9.0,  0.0,  1.0,  2.0,  3.0,  4.0,  0.0,  1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0,  8.0,  9.0,  5.0,
+    6.0,  7.0,  8.0,  9.0,  10.0, 11.0, 12.0, 13.0, 14.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0, 17.0, 18.0,
+    19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 10.0, 11.0, 12.0, 13.0, 14.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0,
+    17.0, 18.0, 19.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0, 20.0, 21.0, 22.0, 23.0, 24.0,
+    25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 20.0, 21.0, 22.0, 23.0, 24.0, 20.0, 21.0, 22.0,
+    23.0, 24.0, 25.0, 26.0, 27.0, 28.0, 29.0, 25.0, 26.0, 27.0, 28.0, 29.0, 30.0, 31.0, 32.0, 33.0, 34.0, 30.0,
+    31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0, 30.0, 31.0, 32.0, 33.0,
+    34.0, 30.0, 31.0, 32.0, 33.0, 34.0, 35.0, 36.0, 37.0, 38.0, 39.0, 35.0, 36.0, 37.0, 38.0, 39.0};
+  size_t output_size = 160;
+  std::vector<float> output(output_size, 0.0);
+
+  thread_num = 4;
+  tid = 0;
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  tid = 1;
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  tid = 2;
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  tid = 3;
+  ResizeNearestNeighbor(input.data(), output.data(), input_shape.data(), output_shape.data(), tid, thread_num);
+  CompareOutputData(output.data(), expect.data(), output_size, err_tol);
+}
+}  // namespace mindspore

mindspore/lite/test/ut/src/runtime/kernel/arm/int8/resize_bilinear_int8_tests.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <iostream>
+#include "include/context.h"
+#include "src/ir/tensor.h"
+#include "common/common_test.h"
+#include "src/common/file_utils.h"
+#include "mindspore/lite/src/kernel_registry.h"
+#include "src/runtime/kernel/arm/nnacl/int8/resize.h"
+#include "src/runtime/kernel/arm/int8/resize_int8.h"
+
+namespace mindspore {
+using mindspore::lite::tensor::QuantArg;
+using mindspore::lite::tensor::Tensor;
+
+class TestResizeBilinearInt8 : public mindspore::Common {
+ public:
+  TestResizeBilinearInt8() = default;
+  void TearDown() override;
+  void Prepare(const std::vector<int> &in_shape, const std::vector<int> &out_shape, int8_t *input_data,
+               int8_t *output_data, const QuantArg quant_in, const QuantArg quant_out, const bool align_corners,
+               const int thread_num);
+  std::vector<lite::tensor::Tensor *> inputs;
+  std::vector<lite::tensor::Tensor *> outputs;
+  ResizeParameter param_ = {};
+  lite::tensor::Tensor in_tensor;
+  lite::tensor::Tensor out_tensor;
+
+  kernel::KernelKey desc_ = {kernel::KERNEL_ARCH::kCPU, kNumberTypeInt8, schema::PrimitiveType_Resize};
+  kernel::KernelCreator creator_ = nullptr;
+  lite::Context ctx_ = lite::Context();
+  kernel::LiteKernel *kernel_ = nullptr;
+  float err_percent_ = 0.2f;
+};
+
+void TestResizeBilinearInt8::TearDown() {
+  in_tensor.SetData(nullptr);
+  out_tensor.SetData(nullptr);
+}
+
+void TestResizeBilinearInt8::Prepare(const std::vector<int> &in_shape, const std::vector<int> &out_shape,
+                                     int8_t *input_data, int8_t *output_data, const mindspore::QuantArg quant_in,
+                                     const mindspore::QuantArg quant_out, const bool align_corners,
+                                     const int thread_num) {
+  in_tensor.set_data_type(kNumberTypeInt8);
+  in_tensor.set_shape(in_shape);
+  in_tensor.SetData(input_data);
+  in_tensor.AddQuantParam(quant_in);
+
+  out_tensor.set_data_type(kNumberTypeInt8);
+  out_tensor.set_shape(out_shape);
+  out_tensor.SetData(output_data);
+  out_tensor.AddQuantParam(quant_out);
+
+  inputs.push_back(&in_tensor);
+  outputs.push_back(&out_tensor);
+
+  param_.method_ = static_cast<int>(schema::ResizeMethod_BILINEAR);
+  param_.new_width_ = out_shape[2];
+  param_.new_height_ = out_shape[1];
+  param_.align_corners_ = align_corners;
+
+  creator_ = lite::KernelRegistry::GetInstance()->GetCreator(desc_);
+
+  ctx_.thread_num_ = thread_num;
+  kernel_ = creator_(inputs, outputs, reinterpret_cast<OpParameter *>(&param_), &ctx_, desc_, nullptr);
+}
+
+TEST_F(TestResizeBilinearInt8, Bilinear0) {
+  int8_t input_data[] = {0, 1, 2, 3};
+  int8_t output_data[16] = {0};
+  std::vector<int> in_shape = {1, 2, 2, 1};
+  std::vector<int> out_shape = {1, 4, 4, 1};
+  const lite::tensor::QuantArg quant_in = {0.005f, 2};
+  const lite::tensor::QuantArg quant_out = {0.008f, 5};
+  bool align_corners = false;
+  int thread_num = 1;
+  int8_t expect[16] = {4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 6, 6, 5, 5, 6, 6};
+
+  Prepare(in_shape, out_shape, input_data, output_data, quant_in, quant_out, align_corners, thread_num);
+  kernel_->Init();
+  kernel_->Run();
+
+  CompareOutputInt8(output_data, expect, 16, err_percent_);
+}
+
+// 2*2*2*5 -> 2*4*4*5
+TEST_F(TestResizeBilinearInt8, Bilinear1) {
+  std::vector<int> in_shape = {2, 2, 2, 5};
+  std::vector<int> out_shape = {2, 4, 4, 5};
+  int8_t input_data[] = {0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 15, 16, 17, 18, 19,
+                         20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39};
+  int8_t output_data[160] = {0};
+  const lite::tensor::QuantArg quant_in = {0.005f, 2};
+  const lite::tensor::QuantArg quant_out = {0.008f, 5};
+  int thread_num = 1;
+  bool align_corners = false;
+  int8_t expect[160] = {4,  4,  5,  6,  6,  5,  6,  7,  7,  8,  7,  8,  8,  9,  9,  7,  8,  8,  9,  9,  7,  8,  8,
+                        9,  9,  8,  9,  10, 10, 11, 10, 11, 11, 12, 13, 10, 11, 11, 12, 13, 10, 11, 11, 12, 13, 12,
+                        12, 13, 13, 14, 13, 14, 14, 15, 16, 13, 14, 14, 15, 16, 10, 11, 11, 12, 13, 12, 12, 13, 13,
+                        14, 13, 14, 14, 15, 16, 13, 14, 14, 15, 16, 16, 17, 18, 18, 19, 18, 18, 19, 20, 20, 19, 20,
+                        21, 21, 22, 19, 20, 21, 21, 22, 19, 20, 21, 21, 22, 21, 22, 22, 23, 23, 23, 23, 24, 24, 25,
+                        23, 23, 24, 24, 25, 23, 23, 24, 24, 25, 24, 25, 25, 26, 27, 26, 26, 27, 28, 28, 26, 26, 27,
+                        28, 28, 23, 23, 24, 24, 25, 24, 25, 25, 26, 27, 26, 26, 27, 28, 28, 26, 26, 27, 28, 28};
+
+  Prepare(in_shape, out_shape, input_data, output_data, quant_in, quant_out, align_corners, thread_num);
+  kernel_->Init();
+  kernel_->Run();
+
+  CompareOutputInt8(output_data, expect, 160, err_percent_);
+}
+
+// 2*2*2*5 -> 2*4*4*5 thread num 2, align corners
+TEST_F(TestResizeBilinearInt8, Bilinear2) {
+  std::vector<int> in_shape = {2, 2, 2, 5};
+  std::vector<int> out_shape = {2, 4, 4, 5};
+  int8_t input_data[] = {0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 15, 16, 17, 18, 19,
+                         20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39};
+  int8_t output_data[160] = {0};
+
+  const lite::tensor::QuantArg quant_in = {0.005f, 2};
+  const lite::tensor::QuantArg quant_out = {0.008f, 5};
+  int thread_num = 2;
+  bool align_corners = true;
+  int8_t expect[160] = {4,  4,  5,  6,  6,  5,  5,  6,  7,  7,  6,  6,  7,  8,  8,  7,  8,  8,  9,  9,  6,  6,  7,
+                        8,  8,  7,  8,  8,  9,  9,  8,  9,  9,  10, 10, 9,  10, 10, 11, 11, 8,  9,  9,  10, 10, 9,
+                        10, 10, 11, 11, 10, 11, 11, 12, 13, 11, 12, 12, 13, 14, 10, 11, 11, 12, 13, 11, 12, 12, 13,
+                        14, 12, 13, 13, 14, 15, 13, 14, 14, 15, 16, 16, 17, 18, 18, 19, 17, 18, 19, 19, 20, 18, 19,
+                        20, 20, 21, 19, 20, 21, 21, 22, 18, 19, 20, 20, 21, 19, 20, 21, 21, 22, 20, 21, 22, 22, 23,
+                        21, 22, 23, 23, 24, 20, 21, 22, 22, 23, 21, 22, 23, 23, 24, 23, 23, 24, 24, 25, 24, 24, 25,
+                        25, 26, 23, 23, 24, 24, 25, 24, 24, 25, 25, 26, 25, 25, 26, 26, 27, 26, 26, 27, 28, 28};
+
+  Prepare(in_shape, out_shape, input_data, output_data, quant_in, quant_out, align_corners, thread_num);
+  kernel_->Init();
+  kernel_->Run();
+
+  CompareOutputInt8(output_data, expect, 160, err_percent_);
+}
+}  // namespace mindspore

mindspore/lite/test/ut/src/runtime/kernel/arm/int8/resize_nearest_neighbor_int8_tests.cc

+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <iostream>
+#include "include/context.h"
+#include "src/ir/tensor.h"
+#include "common/common_test.h"
+#include "mindspore/lite/src/kernel_registry.h"
+#include "src/runtime/kernel/arm/nnacl/int8/resize.h"
+
+namespace mindspore {
+using mindspore::lite::tensor::QuantArg;
+using mindspore::lite::tensor::Tensor;
+
+class TestResizeNearestNeighborInt8 : public mindspore::Common {
+ public:
+  TestResizeNearestNeighborInt8() = default;
+  void Prepare(const std::vector<int> &in_shape, const std::vector<int> &out_shape, int8_t *input_data,
+               int8_t *output_data, const QuantArg quant_in, const QuantArg quant_out, const bool align_corners,
+               const int thread_num);
+  void TearDown() override;
+
+  std::vector<lite::tensor::Tensor *> inputs;
+  std::vector<lite::tensor::Tensor *> outputs;
+  ResizeParameter param_ = {};
+  lite::tensor::Tensor in_tensor;
+  lite::tensor::Tensor out_tensor;
+
+  kernel::KernelKey desc_ = {kernel::KERNEL_ARCH::kCPU, kNumberTypeInt8, schema::PrimitiveType_Resize};
+  kernel::KernelCreator creator_ = nullptr;
+  lite::Context ctx_ = lite::Context();
+  kernel::LiteKernel *kernel_ = nullptr;
+  float err_percent_ = 0.05f;
+};
+
+void TestResizeNearestNeighborInt8::Prepare(const std::vector<int> &in_shape, const std::vector<int> &out_shape,
+                                            int8_t *input_data, int8_t *output_data, const QuantArg quant_in,
+                                            const QuantArg quant_out, const bool align_corners, const int thread_num) {
+  in_tensor.set_data_type(kNumberTypeInt8);
+  in_tensor.set_shape(in_shape);
+  in_tensor.SetData(input_data);
+  in_tensor.AddQuantParam(quant_in);
+
+  out_tensor.set_data_type(kNumberTypeInt8);
+  out_tensor.set_shape(out_shape);
+  out_tensor.SetData(output_data);
+  out_tensor.AddQuantParam(quant_out);
+
+  inputs.push_back(&in_tensor);
+  outputs.push_back(&out_tensor);
+
+  param_.method_ = static_cast<int>(schema::ResizeMethod_NEAREST_NEIGHBOR);
+  param_.new_width_ = out_shape[2];
+  param_.new_height_ = out_shape[1];
+  param_.align_corners_ = align_corners;
+
+  creator_ = lite::KernelRegistry::GetInstance()->GetCreator(desc_);
+
+  ctx_.thread_num_ = thread_num;
+  kernel_ = creator_(inputs, outputs, reinterpret_cast<OpParameter *>(&param_), &ctx_, desc_, nullptr);
+}
+
+void TestResizeNearestNeighborInt8::TearDown() {
+  in_tensor.SetData(nullptr);
+  out_tensor.SetData(nullptr);
+}
+
+// 2*2*1 -> 4*4*1
+TEST_F(TestResizeNearestNeighborInt8, NearestNeighbor0) {
+  std::vector<int> in_shape = {1, 2, 2, 1};
+  std::vector<int> out_shape = {1, 4, 4, 1};
+  QuantArg quant_in = {0.00390625, 2};
+  QuantArg quant_out = {0.015625, 5};
+  int8_t input_data[] = {0, 1, 2, 3};
+  const int out_element_num = 16;
+  int8_t output_data[out_element_num] = {0};
+  int thread_num = 1;
+  int8_t expect[16] = {5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5};
+  err_percent_ = 0.25f;
+
+  Prepare(in_shape, out_shape, input_data, output_data, quant_in, quant_out, false, thread_num);
+  kernel_->Init();
+  kernel_->Run();
+
+  CompareOutputInt8(output_data, expect, 16, err_percent_);
+}
+
+// 2*2*2*5 -> 2*4*4*5
+TEST_F(TestResizeNearestNeighborInt8, NearestNeighbor1) {
+  std::vector<int> in_shape = {2, 2, 2, 5};
+  std::vector<int> out_shape = {2, 4, 4, 5};
+  QuantArg quant_in = {0.00390625, 2};
+  QuantArg quant_out = {0.015625, 5};
+  int8_t input_data[] = {0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 15, 16, 17, 18, 19,
+                         20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39};
+  const int out_element_num = 160;
+  int8_t output_data[out_element_num] = {0};
+  int thread_num = 1;
+  int8_t expect[160] = {5,  5,  5,  5,  6,  5,  5,  5,  5,  6,  6,  6,  6,  7,  7,  6,  6,  6,  7,  7,  5,  5,  5,
+                        5,  6,  5,  5,  5,  5,  6,  6,  6,  6,  7,  7,  6,  6,  6,  7,  7,  7,  7,  8,  8,  8,  7,
+                        7,  8,  8,  8,  8,  9,  9,  9,  9,  8,  9,  9,  9,  9,  7,  7,  8,  8,  8,  7,  7,  8,  8,
+                        8,  8,  9,  9,  9,  9,  8,  9,  9,  9,  9,  10, 10, 10, 10, 11, 10, 10, 10, 10, 11, 11, 11,
+                        11, 12, 12, 11, 11, 11, 12, 12, 10, 10, 10, 10, 11, 10, 10, 10, 10, 11, 11, 11, 11, 12, 12,
+                        11, 11, 11, 12, 12, 12, 12, 13, 13, 13, 12, 12, 13, 13, 13, 13, 14, 14, 14, 14, 13, 14, 14,
+                        14, 14, 12, 12, 13, 13, 13, 12, 12, 13, 13, 13, 13, 14, 14, 14, 14, 13, 14, 14, 14, 14};
+
+  Prepare(in_shape, out_shape, input_data, output_data, quant_in, quant_out, false, thread_num);
+  kernel_->Init();
+  kernel_->Run();
+
+  CompareOutputInt8(output_data, expect, out_element_num, err_percent_);
+}
+
+// 2*2*2*5 -> 2*4*4*5 thread num 2 align_corners
+TEST_F(TestResizeNearestNeighborInt8, NearestNeighbor2) {
+  std::vector<int> in_shape = {2, 2, 2, 5};
+  std::vector<int> out_shape = {2, 4, 4, 5};
+  QuantArg quant_in = {0.00390625, 2};
+  QuantArg quant_out = {0.015625, 5};
+  int8_t input_data[] = {0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 15, 16, 17, 18, 19,
+                         20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39};
+  const int out_element_num = 160;
+  int8_t output_data[out_element_num] = {0};
+  int thread_num = 2;
+  int8_t expect[160] = {
+    5,  5,  5,  5,  6,  5,  5,  5,  5,  6,  6,  6,  6,  7,  7,  6,  6,  6,  7,  7,  5,  5,  5,  5,  6,  5,  5,
+    5,  5,  6,  6,  6,  6,  7,  7,  6,  6,  6,  7,  7,  7,  7,  8,  8,  8,  7,  7,  8,  8,  8,  8,  9,  9,  9,
+    9,  8,  9,  9,  9,  9,  7,  7,  8,  8,  8,  7,  7,  8,  8,  8,  8,  9,  9,  9,  9,  8,  9,  9,  9,  9,  10,
+    10, 10, 10, 11, 10, 10, 10, 10, 11, 11, 11, 11, 12, 12, 11, 11, 11, 12, 12, 10, 10, 10, 10, 11, 10, 10, 10,
+    10, 11, 11, 11, 11, 12, 12, 11, 11, 11, 12, 12, 12, 12, 13, 13, 13, 12, 12, 13, 13, 13, 13, 14, 14, 14, 14,
+    13, 14, 14, 14, 14, 12, 12, 13, 13, 13, 12, 12, 13, 13, 13, 13, 14, 14, 14, 14, 13, 14, 14, 14, 14,
+  };
+
+  Prepare(in_shape, out_shape, input_data, output_data, quant_in, quant_out, true, thread_num);
+  kernel_->Init();
+  kernel_->Run();
+
+  CompareOutputInt8(output_data, expect, out_element_num, err_percent_);
+}
+
+// 2*2*2*5 -> 2*4*4*5 thread num 2, same quant args
+TEST_F(TestResizeNearestNeighborInt8, NearestNeighbor3) {
+  std::vector<int> in_shape = {2, 2, 2, 5};
+  std::vector<int> out_shape = {2, 4, 4, 5};
+  QuantArg quant_in = {0.00390625, 2};
+  QuantArg quant_out = {0.00390625, 2};
+  int8_t input_data[] = {0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 15, 16, 17, 18, 19,
+                         20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39};
+  const int out_element_num = 160;
+  int8_t output_data[out_element_num] = {0};
+  int thread_num = 2;
+  int8_t expect[160] = {0,  1,  2,  3,  4,  0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  5,  6,  7,  8,  9,  0,  1,  2,
+                        3,  4,  0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 10,
+                        11, 12, 13, 14, 15, 16, 17, 18, 19, 15, 16, 17, 18, 19, 10, 11, 12, 13, 14, 10, 11, 12, 13,
+                        14, 15, 16, 17, 18, 19, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 20, 21, 22, 23, 24, 25, 26,
+                        27, 28, 29, 25, 26, 27, 28, 29, 20, 21, 22, 23, 24, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29,
+                        25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 35, 36, 37,
+                        38, 39, 30, 31, 32, 33, 34, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 35, 36, 37, 38, 39};
+
+  Prepare(in_shape, out_shape, input_data, output_data, quant_in, quant_out, false, thread_num);
+  kernel_->Init();
+  kernel_->Run();
+
+  CompareOutputInt8(output_data, expect, out_element_num, err_percent_);
+}
+
+// 2*2*2*5 -> 2*4*4*5 thread num 2 align_corners, same quant args
+TEST_F(TestResizeNearestNeighborInt8, NearestNeighbor4) {
+  std::vector<int> in_shape = {2, 2, 2, 5};
+  std::vector<int> out_shape = {2, 4, 4, 5};
+  QuantArg quant_in = {0.00390625, 2};
+  QuantArg quant_out = {0.00390625, 2};
+  int8_t input_data[] = {0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 15, 16, 17, 18, 19,
+                         20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39};
+  const int out_element_num = 160;
+  int8_t output_data[out_element_num] = {0};
+  int thread_num = 2;
+  int8_t expect[160] = {
+    0,  1,  2,  3,  4,  0,  1,  2,  3,  4,  5,  6,  7,  8,  9,  5,  6,  7,  8,  9,  0,  1,  2,  3,  4,  0,  1,
+    2,  3,  4,  5,  6,  7,  8,  9,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 10, 11, 12, 13, 14, 15, 16, 17, 18,
+    19, 15, 16, 17, 18, 19, 10, 11, 12, 13, 14, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 15, 16, 17, 18, 19, 20,
+    21, 22, 23, 24, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 25, 26, 27, 28, 29, 20, 21, 22, 23, 24, 20, 21, 22,
+    23, 24, 25, 26, 27, 28, 29, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39,
+    35, 36, 37, 38, 39, 30, 31, 32, 33, 34, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 35, 36, 37, 38, 39,
+  };
+
+  Prepare(in_shape, out_shape, input_data, output_data, quant_in, quant_out, true, thread_num);
+  kernel_->Init();
+  kernel_->Run();
+
+  CompareOutputInt8(output_data, expect, out_element_num, err_percent_);
+}
+}  // namespace mindspore